fix: upgrade node-pty to fix macOS PTY exhaustion

[speakman]

Summary

Fixes the macOS PTY exhaustion bug introduced by the node-pty dependency in slopus#553.

• Upgrade node-pty from 1.1.0 to 1.2.0-beta.11 which includes the /dev/ptmx leak fix (microsoft/node-pty#882)
• Fix abort handler to kill the entire process group (-pid, SIGTERM) instead of just the launcher PID (SIGHUP)

Root Cause

node-pty v1.1.0 has an off-by-one bug in pty_posix_spawn() (macOS codepath in src/unix/pty.cc):

// Allocate placeholder fds to keep master fd >= 3
int low_fds[3];
size_t count = 0;

for (; count < 3; count++) {
    low_fds[count] = posix_openpt(O_RDWR);  // allocates a PTY pair!
    if (low_fds[count] >= STDERR_FILENO)
        break;  // count stays at 0 in normal case
}

*master = posix_openpt(O_RDWR);  // the real master

// Cleanup — BUGGY:
for (; count > 0; count--) {
    close(low_fds[count]);  // never executes when count == 0
}

In the normal case (fds 0,1,2 occupied), low_fds[0] gets fd >= 3, the break fires with count = 0, and the cleanup loop count > 0 is immediately false. low_fds[0] is never closed, leaking 1 PTY master fd per pty.spawn() call. With macOS's 511 PTY limit (kern.tty.ptmx_max), this exhausts all PTYs after ~511 mode switches.

The upstream fix (in 1.2.0-beta.9+) changes the cleanup to:

for (size_t i = 0; i <= count; i++) {
    close(low_fds[i]);
}

Test plan

• All 8 claudeLocal.test.ts tests pass
• Verify PTY count stays stable across multiple local↔remote mode switches on macOS
• Verify abort properly terminates both the node launcher and the Claude binary

[mfazekas]

Since this is testing the concept and not the actual implementation, I'm not sure if it should be part of the test suite